add docs

bittersweet1999 · Dec 18, 2023 · 7cbfadf · 7cbfadf
1 parent 9dd44de
commit 7cbfadf
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 2 deletions.
diff --git a/configs/eval_subjective_judge_pandalm.py b/configs/eval_subjective_judge_pandalm.py
@@ -1,5 +1,3 @@
-from os import getenv as gv
-
 from mmengine.config import read_base
 with read_base():
     from .models.qwen.hf_qwen_7b_chat import models as hf_qwen_7b_chat

diff --git a/docs/en/advanced_guides/subjective_evaluation.md b/docs/en/advanced_guides/subjective_evaluation.md
@@ -144,6 +144,13 @@ The `-r` parameter allows the reuse of model inference and GPT-4 evaluation resu
 The response of JudgeLLM will be output to `output/.../results/timestamp/xxmodel/xxdataset/.json`.
 The evaluation report will be output to `output/.../summary/timestamp/report.csv`.
 
+Opencompass has supported lots of JudgeLLM, actually, you can take any model as JudgeLLM in opencompass configs.
+And we list the popular open-source JudgeLLM here:
+
+1. Auto-J, refer to `configs/models/judge_llm/auto_j`
+2. JudgeLM, refer to `configs/models/judge_llm/judgelm`
+3. PandaLM, refer to `configs/models/judge_llm/pandalm`
+
 ## Practice: AlignBench Evaluation
 
 ### Dataset

diff --git a/docs/zh_cn/advanced_guides/subjective_evaluation.md b/docs/zh_cn/advanced_guides/subjective_evaluation.md
@@ -142,6 +142,13 @@ python run.py configs/eval_subjective_score.py -r
 JudgeLLM的评测回复会保存在 `output/.../results/timestamp/xxmodel/xxdataset/.json`
 评测报告则会输出到 `output/.../summary/timestamp/report.csv`。
 
+Opencompass 已经支持了很多的JudgeLLM，实际上，你可以将Opencompass中所支持的所有模型都当作JudgeLLM使用。
+我们列出目前比较流行的开源JudgeLLM：
+
+1. Auto-J，请参考 `configs/models/judge_llm/auto_j`
+2. JudgeLM，请参考 `configs/models/judge_llm/judgelm`
+3. PandaLM，请参考 `configs/models/judge_llm/pandalm`
+
 ## 实战：AlignBench 主观评测
 
 ### 数据集准备